Growth Models II
Emorie D Beck
Workspace
Packages
library(psych)
library(broom)
library(lme4)
library(knitr)
library(kableExtra)
library(plyr)
library(tidyverse)Data
This week, our data are going to (continue to) come from the German Socioeconomic Panel Study (GSOEP). The GSOEP is a longitudinal study of adults in German housesholds. The study has a broad range of variables, but for our purposes we’re just going to use personality ratings, life events, age, gender, and general life satisfaction from 2005 to 2015. We’ll use life satisfaction from 2005 this week, but keep the other years for future use when we do time-varying moderators. We’ll use life events similarly to how we did for logistic regression, except this time as a moderator variable.
We need to extend our codebook. To create it, go to https://data.soep.de/soep-core# and use the search feature to find the variables you need or https://data.soep.de/soep-long/topics/ where you can scroll through topics (this may be easier for finding the personality variables). Use your codebook from last week, and add the additional variables.
Each year has several different files. Thankfully, for our purposes, we just need one file for each year. The first part of that file name indexes which wave it is. Waves are labeled a (1985) to bf (2015). Once the waves hit z, they start over at “ba”. The second piece of the filename indexes which type of file it is. We need the “p” files, which stand for person. So, for instance, 2005 is “vp.sav”.
There are different ways to load it in, but I would recommend using some form of loop, which should do the following:
1. read in the file for a specific year (e.g. using haven::read_sav()). 2. pull the variables from the codebook from that year (e.g. using select()). - NOTE: you should pull certain variables, like the person and household IDs for every year.
3. rename those variables in wide format.
4. add a column to the data for that year that indexes what year the observation is.
5. merge the data from that year with previous years.
For help with this, see https://emoriebeck.github.io/R-tutorials/purrr/. I’ll give you a purrr solution later in the week.
Once you’ve got the codebook, we should be ready to go.
# wd <- "https://github.com/emoriebeck/R-tutorials/blob/master/RA_Files/Week_7_Growth_Models_II"
wd <- "~/Documents/Github/R-tutorials/RA_Files/Week_7_Growth_Models_II"
# load your codebook
# destfile <- "Codebook_EDB.xlsx"
# curl::curl_download(sprintf("%s/Codebook_EDB.xlsx?raw=true", wd), destfile)
# codebook <- readxl::read_excel(destfile) %>%
codebook <- readxl::read_excel(sprintf("%s/Codebook_EDB.xlsx", wd)) %>%
mutate(Item = stringr::str_to_lower(Item))# data_path <- "https://github.com/emoriebeck/R-tutorials/blob/master/RA_Files/Week_5_Logistic_Regression"
data_path <- "~/Documents/Github/R-tutorials/RA_Files/Week_5_Logistic_Regression"
all.old.cols <- (codebook %>% filter(class == "proc" & Year == 0))$Item
all.new.cols <- (codebook %>% filter(class == "proc" & Year == "0"))$new_name
# create short function to read in separate files for each wave
read_fun <- function(file, year){
print(year)
old.names <- (codebook %>% filter(Year == year))$Item
new.names <- (codebook %>% filter(Year == year))$new_name
# z <- read_csv(url(sprintf("%s/data/%sp.csv?raw=true", data_path, file))) %>%
z <- read.csv(sprintf("%s/data/%sp.csv", data_path, file)) %>%
select(one_of(all.old.cols), one_of(old.names)) %>%
setNames(c(all.new.cols, new.names))
}
# you need letters, not numbers to index different data files.
# but years will be more useful to index your codebook, so we'll
# put both in our starting data frame. I've filled out this part.
# Now you just need to figure out how use that to load the files
# and get the correct variables (one's that repeat year to year)
dat <- tibble(
Year = as.character(seq(2005, 2015,1)),
file = c(letters[22:26], paste("b", letters[1:6], sep = ""))) %>%
mutate(data = map2(file, Year, read_fun)) %>%
unnest(data)## [1] "2005"
## [1] "2006"
## [1] "2007"
## [1] "2008"
## [1] "2009"
## [1] "2010"
## [1] "2011"
## [1] "2012"
## [1] "2013"
## [1] "2014"
## [1] "2015"Descriptives
Because our data are now longitudinal, we need to split our descriptives by year. Try doing this using the describeBy() in the psych package.
# run the descriptives and check variable ranges
describeBy(dat, dat$Year)##
## Descriptive statistics by group
## group: 2005
## vars n mean sd median trimmed
## Year* 1 10451 2005.00 0.00 2005 2005.00
## file* 2 10451 NaN NA NA NaN
## PROC_SID 3 10451 3006396.71 2420592.24 2727001 2737008.54
## PROC_household 4 10451 297467.55 245527.78 273139 270013.72
## Dem_DOB 5 10451 1957.45 17.53 1958 1957.78
## Dem_Sex 6 10451 1.52 0.50 2 1.53
## BF_C1 7 10451 6.12 1.19 6 6.33
## BF_E1 8 10451 5.45 1.42 6 5.62
## BF_A1 9 10451 2.90 1.70 3 2.76
## BF_O1 10 10451 4.52 1.53 5 4.60
## BF_N1 11 10451 4.67 1.72 5 4.78
## BF_A2 12 10451 5.48 1.40 6 5.66
## BF_C2 13 10451 2.22 1.55 2 1.96
## BF_E2 14 10451 5.05 1.50 5 5.18
## BF_O2 15 10451 4.09 1.88 4 4.14
## BF_N2 16 10451 3.67 1.76 4 3.66
## BF_C3 17 10451 5.71 1.29 6 5.90
## BF_E3 18 10451 4.08 1.68 4 4.12
## BF_A3 19 10451 5.75 1.20 6 5.90
## BF_O3 20 10451 4.77 1.60 5 4.89
## BF_N3 21 10451 4.51 1.57 5 4.58
## LE_ChldBrth 22 10451 -1.93 0.45 -2 -2.00
## LE_ChldMvOut 23 10451 -1.93 0.46 -2 -2.00
## LE_Divorce 24 10451 -1.98 0.21 -2 -2.00
## LE_DadDied 25 10451 -1.97 0.31 -2 -2.00
## LE_Married 26 10451 -1.95 0.37 -2 -2.00
## LE_MomDied 27 10451 -1.97 0.30 -2 -2.00
## LE_MoveIn 28 10451 -1.95 0.40 -2 -2.00
## LE_PartDied 29 10451 -1.99 0.20 -2 -2.00
## LE_SepPart 30 10451 -1.95 0.38 -2 -2.00
## Psych_LifeSat 31 10451 6.92 1.89 7 7.11
## LE_NewPart 32 0 NaN NA NA NaN
## mad min max range skew kurtosis se
## Year* 0.00 2005 2005 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 3246593.03 901 8263202 8262301 0.74 -0.35 23677.87
## PROC_household 340195.91 94 824984 824890 0.70 -0.41 2401.72
## Dem_DOB 20.76 1909 1988 79 -0.15 -0.80 0.17
## Dem_Sex 0.00 1 2 1 -0.10 -1.99 0.00
## BF_C1 1.48 -1 7 8 -2.45 9.39 0.01
## BF_E1 1.48 -1 7 8 -1.08 1.71 0.01
## BF_A1 1.48 -1 7 8 0.50 -0.65 0.02
## BF_O1 1.48 -1 7 8 -0.60 0.62 0.02
## BF_N1 1.48 -1 7 8 -0.48 -0.33 0.02
## BF_A2 1.48 -1 7 8 -1.23 2.20 0.01
## BF_C2 1.48 -1 7 8 1.16 0.66 0.02
## BF_E2 1.48 -1 7 8 -0.74 0.59 0.01
## BF_O2 1.48 -1 7 8 -0.24 -0.72 0.02
## BF_N2 1.48 -1 7 8 0.07 -0.80 0.02
## BF_C3 1.48 -1 7 8 -1.87 5.92 0.01
## BF_E3 1.48 -1 7 8 -0.24 -0.53 0.02
## BF_A3 1.48 -1 7 8 -1.48 4.22 0.01
## BF_O3 1.48 -1 7 8 -0.62 0.28 0.02
## BF_N3 1.48 -1 7 8 -0.48 0.04 0.02
## LE_ChldBrth 0.00 -2 1 3 6.30 37.67 0.00
## LE_ChldMvOut 0.00 -2 1 3 6.23 36.82 0.00
## LE_Divorce 0.00 -2 1 3 13.93 192.16 0.00
## LE_DadDied 0.00 -2 1 3 9.46 87.48 0.00
## LE_Married 0.00 -2 1 3 7.79 58.73 0.00
## LE_MomDied 0.00 -2 1 3 9.87 95.48 0.00
## LE_MoveIn 0.00 -2 1 3 7.23 50.30 0.00
## LE_PartDied 0.00 -2 1 3 14.50 208.25 0.00
## LE_SepPart 0.00 -2 1 3 7.67 56.85 0.00
## Psych_LifeSat 1.48 -1 10 11 -1.04 1.38 0.02
## LE_NewPart NA Inf -Inf -Inf NA NA NA
## --------------------------------------------------------
## group: 2006
## vars n mean sd median trimmed
## Year* 1 11085 2006.00 0.00 2006 2006.00
## file* 2 11085 NaN NA NA NaN
## PROC_SID 3 11085 3059412.03 2281556.03 2849801 2811416.13
## PROC_household 4 11085 303086.61 231649.53 285170 277923.72
## Dem_DOB 5 11085 1957.39 17.34 1958 1957.63
## Dem_Sex 6 11085 1.53 0.50 2 1.53
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 11085 -1.91 0.50 -2 -2.00
## LE_ChldMvOut 23 11085 -1.90 0.53 -2 -2.00
## LE_Divorce 24 11085 -1.98 0.23 -2 -2.00
## LE_DadDied 25 11085 -1.97 0.31 -2 -2.00
## LE_Married 26 11085 -1.95 0.37 -2 -2.00
## LE_MomDied 27 11085 -1.96 0.33 -2 -2.00
## LE_MoveIn 28 11085 -1.94 0.43 -2 -2.00
## LE_PartDied 29 11085 -1.99 0.20 -2 -2.00
## LE_SepPart 30 11085 -1.95 0.38 -2 -2.00
## Psych_LifeSat 31 11085 6.91 1.85 7 7.08
## LE_NewPart 32 0 NaN NA NA NaN
## mad min max range skew kurtosis se
## Year* 0.00 2006 2006 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 2637099.14 901 8268502 8267601 0.73 -0.07 21670.22
## PROC_household 324072.64 94 824984 824890 0.69 -0.14 2200.21
## Dem_DOB 19.27 1909 1989 80 -0.12 -0.82 0.16
## Dem_Sex 0.00 1 2 1 -0.10 -1.99 0.00
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -2 1 3 5.65 29.88 0.00
## LE_ChldMvOut 0.00 -2 1 3 5.32 26.34 0.01
## LE_Divorce 0.00 -2 1 3 12.74 160.42 0.00
## LE_DadDied 0.00 -2 1 3 9.45 87.37 0.00
## LE_Married 0.00 -2 1 3 7.94 60.99 0.00
## LE_MomDied 0.00 -2 1 3 8.96 78.34 0.00
## LE_MoveIn 0.00 -2 1 3 6.66 42.39 0.00
## LE_PartDied 0.00 -2 1 3 14.94 221.19 0.00
## LE_SepPart 0.00 -2 1 3 7.61 55.91 0.00
## Psych_LifeSat 1.48 -1 10 11 -1.03 1.63 0.02
## LE_NewPart NA Inf -Inf -Inf NA NA NA
## --------------------------------------------------------
## group: 2007
## vars n mean sd median trimmed
## Year* 1 10438 2007.00 0.00 2007 2007.00
## file* 2 10438 NaN NA NA NaN
## PROC_SID 3 10438 3053902.85 2303829.10 2829602 2803303.50
## PROC_household 4 10438 302841.40 234249.72 283339 277502.91
## Dem_DOB 5 10438 1958.09 17.41 1959 1958.30
## Dem_Sex 6 10438 1.53 0.50 2 1.53
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 10438 -1.93 0.44 -2 -2.00
## LE_ChldMvOut 23 10438 -1.91 0.52 -2 -2.00
## LE_Divorce 24 10438 -1.98 0.22 -2 -2.00
## LE_DadDied 25 10438 -1.97 0.30 -2 -2.00
## LE_Married 26 10438 -1.95 0.40 -2 -2.00
## LE_MomDied 27 10438 -1.97 0.30 -2 -2.00
## LE_MoveIn 28 10438 -1.92 0.47 -2 -2.00
## LE_PartDied 29 10438 -1.99 0.18 -2 -2.00
## LE_SepPart 30 10438 -1.95 0.39 -2 -2.00
## Psych_LifeSat 31 10438 6.93 1.84 7 7.11
## LE_NewPart 32 0 NaN NA NA NaN
## mad min max range skew kurtosis se
## Year* 0.00 2007 2007 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 2770608.01 901 8271402 8270501 0.73 -0.13 22549.74
## PROC_household 329309.18 94 824984 824890 0.68 -0.21 2292.82
## Dem_DOB 19.27 1909 1989 80 -0.11 -0.83 0.17
## Dem_Sex 0.00 1 2 1 -0.10 -1.99 0.00
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -2 1 3 6.48 40.01 0.00
## LE_ChldMvOut 0.00 -2 1 3 5.43 27.44 0.01
## LE_Divorce 0.00 -2 1 3 13.67 184.75 0.00
## LE_DadDied 0.00 -2 1 3 9.63 90.75 0.00
## LE_Married 0.00 -2 1 3 7.29 51.13 0.00
## LE_MomDied 0.00 -2 1 3 9.77 93.46 0.00
## LE_MoveIn 0.00 -2 1 3 6.05 34.56 0.00
## LE_PartDied 0.00 -2 1 3 16.94 284.89 0.00
## LE_SepPart 0.00 -2 1 3 7.53 54.65 0.00
## Psych_LifeSat 1.48 -1 10 11 -1.06 1.57 0.02
## LE_NewPart NA Inf -Inf -Inf NA NA NA
## --------------------------------------------------------
## group: 2008
## vars n mean sd median trimmed
## Year* 1 9826 2008.00 0.00 2008 2008.00
## file* 2 9826 NaN NA NA NaN
## PROC_SID 3 9826 3054938.00 2305363.46 2829602 2804337.97
## PROC_household 4 9826 302745.32 234444.13 283347 277365.25
## Dem_DOB 5 9826 1958.60 17.53 1959 1958.74
## Dem_Sex 6 9826 1.52 0.50 2 1.53
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 9826 -1.93 0.44 -2 -2.00
## LE_ChldMvOut 23 9826 -1.92 0.47 -2 -2.00
## LE_Divorce 24 9826 -1.99 0.20 -2 -2.00
## LE_DadDied 25 9826 -1.97 0.30 -2 -2.00
## LE_Married 26 9826 -1.96 0.36 -2 -2.00
## LE_MomDied 27 9826 -1.97 0.30 -2 -2.00
## LE_MoveIn 28 9826 -1.95 0.39 -2 -2.00
## LE_PartDied 29 9826 -1.99 0.21 -2 -2.00
## LE_SepPart 30 9826 -1.95 0.39 -2 -2.00
## Psych_LifeSat 31 9826 6.98 1.79 7 7.14
## LE_NewPart 32 0 NaN NA NA NaN
## mad min max range skew kurtosis se
## Year* 0.00 2008 2008 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 2783729.02 901 8276202 8275301 0.73 -0.14 23256.86
## PROC_household 329567.15 94 824984 824890 0.67 -0.22 2365.11
## Dem_DOB 19.27 1909 1990 81 -0.07 -0.84 0.18
## Dem_Sex 0.00 1 2 1 -0.09 -1.99 0.01
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -2 1 3 6.54 40.72 0.00
## LE_ChldMvOut 0.00 -2 1 3 6.00 34.01 0.00
## LE_Divorce 0.00 -2 1 3 14.51 208.57 0.00
## LE_DadDied 0.00 -2 1 3 9.71 92.28 0.00
## LE_Married 0.00 -2 1 3 7.99 61.85 0.00
## LE_MomDied 0.00 -2 1 3 9.97 97.34 0.00
## LE_MoveIn 0.00 -2 1 3 7.43 53.15 0.00
## LE_PartDied 0.00 -2 1 3 14.05 195.50 0.00
## LE_SepPart 0.00 -2 1 3 7.40 52.81 0.00
## Psych_LifeSat 1.48 -1 10 11 -1.03 1.50 0.02
## LE_NewPart NA Inf -Inf -Inf NA NA NA
## --------------------------------------------------------
## group: 2009
## vars n mean sd median trimmed
## Year* 1 10327 2009.00 0.00 2009 2009.00
## file* 2 10327 NaN NA NA NaN
## PROC_SID 3 10327 3746217.99 2832106.78 2981502 3564937.58
## PROC_household 4 10327 372040.85 286840.94 298352 353663.26
## Dem_DOB 5 10327 1959.03 17.62 1959 1959.10
## Dem_Sex 6 10327 1.52 0.50 2 1.53
## BF_C1 7 10327 6.09 1.21 6 6.31
## BF_E1 8 10327 5.39 1.44 6 5.56
## BF_A1 9 10327 2.99 1.66 3 2.88
## BF_O1 10 10327 4.46 1.52 4 4.52
## BF_N1 11 10327 4.32 1.72 4 4.34
## BF_A2 12 10327 5.33 1.45 6 5.51
## BF_C2 13 10327 2.39 1.60 2 2.17
## BF_E2 14 10327 4.93 1.55 5 5.04
## BF_O2 15 10327 3.96 1.90 4 3.97
## BF_N2 16 10327 3.57 1.70 4 3.54
## BF_C3 17 10327 5.69 1.30 6 5.87
## BF_E3 18 10327 4.05 1.65 4 4.09
## BF_A3 19 10327 5.65 1.24 6 5.80
## BF_O3 20 10327 4.65 1.61 5 4.74
## BF_N3 21 10327 4.44 1.55 5 4.52
## LE_ChldBrth 22 10327 -1.92 0.49 -2 -2.00
## LE_ChldMvOut 23 10327 -1.93 0.45 -2 -2.00
## LE_Divorce 24 10327 -1.99 0.20 -2 -2.00
## LE_DadDied 25 10327 -1.97 0.29 -2 -2.00
## LE_Married 26 10327 -1.95 0.37 -2 -2.00
## LE_MomDied 27 10327 -1.97 0.32 -2 -2.00
## LE_MoveIn 28 10327 -1.94 0.43 -2 -2.00
## LE_PartDied 29 10327 -1.98 0.23 -2 -2.00
## LE_SepPart 30 10327 -1.95 0.38 -2 -2.00
## Psych_LifeSat 31 10327 6.96 1.81 7 7.12
## LE_NewPart 32 0 NaN NA NA NaN
## mad min max range skew kurtosis se
## Year* 0.00 2009 2009 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 3127989.48 901 8999901 8999000 0.57 -0.87 27869.07
## PROC_household 320260.87 94 899992 899898 0.53 -0.89 2822.63
## Dem_DOB 19.27 1909 1992 83 -0.04 -0.85 0.17
## Dem_Sex 0.00 1 2 1 -0.10 -1.99 0.00
## BF_C1 1.48 -1 7 8 -2.37 8.75 0.01
## BF_E1 1.48 -1 7 8 -1.00 1.42 0.01
## BF_A1 1.48 -1 7 8 0.45 -0.59 0.02
## BF_O1 1.48 -1 7 8 -0.48 0.46 0.01
## BF_N1 1.48 -1 7 8 -0.23 -0.62 0.02
## BF_A2 1.48 -1 7 8 -1.13 1.76 0.01
## BF_C2 1.48 -1 7 8 0.97 0.28 0.02
## BF_E2 1.48 -1 7 8 -0.70 0.59 0.02
## BF_O2 1.48 -1 7 8 -0.12 -0.82 0.02
## BF_N2 1.48 -1 7 8 0.13 -0.69 0.02
## BF_C3 1.48 -1 7 8 -1.80 5.45 0.01
## BF_E3 1.48 -1 7 8 -0.26 -0.38 0.02
## BF_A3 1.48 -1 7 8 -1.52 4.45 0.01
## BF_O3 1.48 -1 7 8 -0.48 0.00 0.02
## BF_N3 1.48 -1 7 8 -0.46 0.01 0.02
## LE_ChldBrth 0.00 -2 1 3 5.82 31.90 0.00
## LE_ChldMvOut 0.00 -2 1 3 6.29 37.51 0.00
## LE_Divorce 0.00 -2 1 3 14.88 219.46 0.00
## LE_DadDied 0.00 -2 1 3 10.06 99.30 0.00
## LE_Married 0.00 -2 1 3 7.92 60.78 0.00
## LE_MomDied 0.00 -2 1 3 9.11 81.05 0.00
## LE_MoveIn 0.00 -2 1 3 6.66 42.38 0.00
## LE_PartDied 0.00 -2 1 3 13.11 170.01 0.00
## LE_SepPart 0.00 -2 1 3 7.62 56.11 0.00
## Psych_LifeSat 1.48 -1 10 11 -0.97 1.37 0.02
## LE_NewPart NA Inf -Inf -Inf NA NA NA
## --------------------------------------------------------
## group: 2010
## vars n mean sd median trimmed
## Year* 1 13230 2010.00 0.00 2010 2010.00
## file* 2 13230 NaN NA NA NaN
## PROC_SID 3 13230 8687797.90 8118552.20 5076254 8161911.97
## PROC_household 4 13230 867077.29 813872.45 507857 814715.84
## Dem_DOB 5 13230 1379.80 895.97 1948 1477.01
## Dem_Sex 6 13230 -0.40 3.00 1 -0.13
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 13230 -2.84 1.45 -2 -2.73
## LE_ChldMvOut 23 13230 -2.83 1.46 -2 -2.73
## LE_Divorce 24 13230 -2.87 1.39 -2 -2.73
## LE_DadDied 25 13230 -2.86 1.41 -2 -2.73
## LE_Married 26 13230 -2.85 1.42 -2 -2.73
## LE_MomDied 27 13230 -2.86 1.41 -2 -2.73
## LE_MoveIn 28 13230 -2.84 1.45 -2 -2.73
## LE_PartDied 29 13230 -2.87 1.39 -2 -2.73
## LE_SepPart 30 13230 -2.86 1.42 -2 -2.73
## Psych_LifeSat 31 13230 7.24 1.76 8 7.40
## LE_NewPart 32 0 NaN NA NA NaN
## mad min max range skew kurtosis se
## Year* 0.00 2010 2010 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 5537436.87 901 21536602 21535701 0.66 -1.29 70582.76
## PROC_household 549392.26 94 2153664 2153570 0.65 -1.29 7075.81
## Dem_DOB 34.10 -5 1992 1997 -0.90 -1.19 7.79
## Dem_Sex 1.48 -5 2 7 -0.84 -1.21 0.03
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -5 1 6 -0.55 -0.66 0.01
## LE_ChldMvOut 0.00 -5 1 6 -0.50 -0.61 0.01
## LE_Divorce 0.00 -5 1 6 -0.79 -1.02 0.01
## LE_DadDied 0.00 -5 1 6 -0.72 -0.91 0.01
## LE_Married 0.00 -5 1 6 -0.65 -0.80 0.01
## LE_MomDied 0.00 -5 1 6 -0.72 -0.91 0.01
## LE_MoveIn 0.00 -5 1 6 -0.55 -0.67 0.01
## LE_PartDied 0.00 -5 1 6 -0.79 -1.01 0.01
## LE_SepPart 0.00 -5 1 6 -0.67 -0.83 0.01
## Psych_LifeSat 1.48 -1 10 11 -1.08 1.71 0.02
## LE_NewPart NA Inf -Inf -Inf NA NA NA
## --------------------------------------------------------
## group: 2011
## vars n mean sd median trimmed
## Year* 1 14347 2011.00 0.00 2011 2011.00
## file* 2 14347 NaN NA NA NaN
## PROC_SID 3 14347 12737750.79 11308717.31 7235502 12050013.39
## PROC_household 4 14347 1272299.99 1132701.85 723347 1203922.14
## Dem_DOB 5 14347 1428.98 872.61 1950 1538.29
## Dem_Sex 6 14347 1.55 0.50 2 1.56
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 14347 -2.00 0.93 -2 -2.00
## LE_ChldMvOut 23 14347 -2.08 0.80 -2 -2.00
## LE_Divorce 24 14347 -2.13 0.68 -2 -2.00
## LE_DadDied 25 14347 -2.11 0.72 -2 -2.00
## LE_Married 26 14347 -2.10 0.75 -2 -2.00
## LE_MomDied 27 14347 -2.11 0.72 -2 -2.00
## LE_MoveIn 28 14347 -2.10 0.76 -2 -2.00
## LE_PartDied 29 14347 -2.13 0.67 -2 -2.00
## LE_SepPart 30 14347 -2.10 0.74 -2 -2.00
## Psych_LifeSat 31 14347 6.11 3.86 7 7.00
## LE_NewPart 32 14347 -2.07 0.82 -2 -2.00
## mad min max range skew kurtosis se
## Year* 0.00 2011 2011 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 10362634.18 901 30979502 30978601 0.39 -1.45 94413.22
## PROC_household 1044902.38 94 3097950 3097856 0.39 -1.45 9456.60
## Dem_DOB 29.65 -5 1993 1998 -1.03 -0.93 7.29
## Dem_Sex 0.00 1 2 1 -0.18 -1.97 0.00
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -5 1 6 -0.04 7.49 0.01
## LE_ChldMvOut 0.00 -5 1 6 -1.04 10.73 0.01
## LE_Divorce 0.00 -5 1 6 -3.13 14.48 0.01
## LE_DadDied 0.00 -5 1 6 -2.20 13.11 0.01
## LE_Married 0.00 -5 1 6 -1.70 12.21 0.01
## LE_MomDied 0.00 -5 1 6 -2.22 13.15 0.01
## LE_MoveIn 0.00 -5 1 6 -1.65 12.10 0.01
## LE_PartDied 0.00 -5 1 6 -3.30 14.70 0.01
## LE_SepPart 0.00 -5 1 6 -1.85 12.49 0.01
## Psych_LifeSat 1.48 -5 10 15 -2.04 3.32 0.03
## LE_NewPart 0.00 -5 1 6 -0.86 10.27 0.01
## --------------------------------------------------------
## group: 2012
## vars n mean sd median trimmed
## Year* 1 13996 2012.00 0.00 2012 2012.00
## file* 2 13996 NaN NA NA NaN
## PROC_SID 3 13996 14264024.93 11848265.32 8250354 13898035.71
## PROC_household 4 13996 1425277.75 1185855.89 824507 1389517.26
## Dem_DOB 5 13996 1449.04 862.37 1951 1563.27
## Dem_Sex 6 13996 0.25 2.63 1 0.69
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 13996 -1.93 0.45 -2 -2.00
## LE_ChldMvOut 23 13996 -1.92 0.48 -2 -2.00
## LE_Divorce 24 13996 -1.98 0.22 -2 -2.00
## LE_DadDied 25 13996 -1.97 0.30 -2 -2.00
## LE_Married 26 13996 -1.94 0.42 -2 -2.00
## LE_MomDied 27 13996 -1.97 0.30 -2 -2.00
## LE_MoveIn 28 13996 -1.95 0.37 -2 -2.00
## LE_PartDied 29 13996 -1.99 0.17 -2 -2.00
## LE_SepPart 30 13996 -1.95 0.40 -2 -2.00
## Psych_LifeSat 31 13996 218.73 211.42 200 204.37
## LE_NewPart 32 13996 -1.91 0.50 -2 -2.00
## mad min max range skew kurtosis se
## Year* 0.00 2012 2012 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 12205433.34 901 35008702 35007801 0.22 -1.59 100150.43
## PROC_household 1219978.17 94 3143391 3143297 0.21 -1.59 10023.74
## Dem_DOB 29.65 -5 1994 1999 -1.09 -0.81 7.29
## Dem_Sex 1.48 -5 2 7 -1.42 0.19 0.02
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -2 1 3 6.35 38.35 0.00
## LE_ChldMvOut 0.00 -2 1 3 5.90 32.85 0.00
## LE_Divorce 0.00 -2 1 3 13.28 174.42 0.00
## LE_DadDied 0.00 -2 1 3 9.67 91.52 0.00
## LE_Married 0.00 -2 1 3 6.83 44.64 0.00
## LE_MomDied 0.00 -2 1 3 9.81 94.26 0.00
## LE_MoveIn 0.00 -2 1 3 7.82 59.21 0.00
## LE_PartDied 0.00 -2 1 3 17.55 305.98 0.00
## LE_SepPart 0.00 -2 1 3 7.26 50.77 0.00
## Psych_LifeSat 299.49 -3 980 983 0.25 -1.36 1.79
## LE_NewPart 0.00 -2 1 3 5.62 29.58 0.00
## --------------------------------------------------------
## group: 2013
## vars n mean sd median trimmed
## Year* 1 15544 2013.00 0.00 2013 2013.00
## file* 2 15544 NaN NA NA NaN
## PROC_SID 3 15544 16994593.68 12707407.51 20219202 17089462.50
## PROC_household 4 15544 1698114.62 1270928.67 2021768 1709304.13
## Dem_DOB 5 15544 1204.85 956.41 1943 1258.13
## Dem_Sex 6 15544 -0.96 3.20 1 -0.83
## BF_C1 7 15544 1.84 5.48 5 2.05
## BF_E1 8 15544 1.48 5.23 4 1.60
## BF_A1 9 15544 -0.07 4.11 1 -0.17
## BF_O1 10 15544 0.93 4.83 4 0.97
## BF_N1 11 15544 0.65 4.66 2 0.62
## BF_A2 12 15544 1.40 5.18 4 1.50
## BF_C2 13 15544 -0.44 3.82 1 -0.58
## BF_E2 14 15544 1.23 5.06 4 1.29
## BF_O2 15 15544 0.68 4.73 2 0.63
## BF_N2 16 15544 0.30 4.40 2 0.22
## BF_C3 17 15544 1.61 5.31 5 1.77
## BF_E3 18 15544 0.59 4.60 2 0.57
## BF_A3 19 15544 1.63 5.32 5 1.79
## BF_O3 20 15544 1.03 4.92 4 1.05
## BF_N3 21 15544 0.89 4.80 3 0.92
## LE_ChldBrth 22 15544 -2.43 1.19 -2 -2.22
## LE_ChldMvOut 23 15544 -2.41 1.22 -2 -2.22
## LE_Divorce 24 15544 -2.47 1.12 -2 -2.22
## LE_DadDied 25 15544 -2.45 1.15 -2 -2.22
## LE_Married 26 15544 -2.42 1.20 -2 -2.22
## LE_MomDied 27 15544 -2.45 1.15 -2 -2.22
## LE_MoveIn 28 15544 -2.44 1.16 -2 -2.22
## LE_PartDied 29 15544 -2.47 1.12 -2 -2.22
## LE_SepPart 30 15544 -2.43 1.18 -2 -2.22
## Psych_LifeSat 31 15544 7.30 1.77 8 7.46
## LE_NewPart 32 15544 -2.39 1.24 -2 -2.22
## mad min max range skew kurtosis se
## Year* 0.00 2013 2013 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 17929453.19 901 35022002 35021101 -0.06 -1.68 101923.79
## PROC_household 1792455.99 94 3343277 3343183 -0.06 -1.68 10193.89
## Dem_DOB 53.37 -5 1995 2000 -0.47 -1.77 7.67
## Dem_Sex 1.48 -5 2 7 -0.44 -1.74 0.03
## BF_C1 2.97 -5 7 12 -0.41 -1.77 0.04
## BF_E1 4.45 -5 7 12 -0.36 -1.75 0.04
## BF_A1 5.93 -5 7 12 -0.15 -1.58 0.03
## BF_O1 2.97 -5 7 12 -0.31 -1.72 0.04
## BF_N1 5.93 -5 7 12 -0.23 -1.67 0.04
## BF_A2 4.45 -5 7 12 -0.35 -1.75 0.04
## BF_C2 4.45 -5 7 12 -0.11 -1.47 0.03
## BF_E2 4.45 -5 7 12 -0.33 -1.74 0.04
## BF_O2 5.93 -5 7 12 -0.20 -1.67 0.04
## BF_N2 4.45 -5 7 12 -0.20 -1.64 0.04
## BF_C3 2.97 -5 7 12 -0.39 -1.76 0.04
## BF_E3 5.93 -5 7 12 -0.25 -1.69 0.04
## BF_A3 2.97 -5 7 12 -0.40 -1.76 0.04
## BF_O3 2.97 -5 7 12 -0.30 -1.72 0.04
## BF_N3 4.45 -5 7 12 -0.30 -1.72 0.04
## LE_ChldBrth 0.00 -5 1 6 -1.17 1.72 0.01
## LE_ChldMvOut 0.00 -5 1 6 -0.99 1.73 0.01
## LE_Divorce 0.00 -5 1 6 -1.67 1.55 0.01
## LE_DadDied 0.00 -5 1 6 -1.46 1.65 0.01
## LE_Married 0.00 -5 1 6 -1.10 1.73 0.01
## LE_MomDied 0.00 -5 1 6 -1.46 1.65 0.01
## LE_MoveIn 0.00 -5 1 6 -1.34 1.69 0.01
## LE_PartDied 0.00 -5 1 6 -1.71 1.53 0.01
## LE_SepPart 0.00 -5 1 6 -1.21 1.72 0.01
## Psych_LifeSat 1.48 -1 10 11 -1.12 1.93 0.01
## LE_NewPart 0.00 -5 1 6 -0.86 1.71 0.01
## --------------------------------------------------------
## group: 2014
## vars n mean sd median trimmed
## Year* 1 13777 2014.00 0.00 2014 2014.00
## file* 2 13777 NaN NA NA NaN
## PROC_SID 3 13777 16326819.90 12763111.56 20158602 16264090.54
## PROC_household 4 13777 1631250.64 1275101.57 2015750 1627050.99
## Dem_DOB 5 13777 1692.96 679.34 1962 1867.69
## Dem_Sex 6 13777 1.54 0.50 2 1.55
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 13777 -1.94 0.42 -2 -2.00
## LE_ChldMvOut 23 13777 -1.92 0.49 -2 -2.00
## LE_Divorce 24 13777 -1.99 0.21 -2 -2.00
## LE_DadDied 25 13777 -1.97 0.32 -2 -2.00
## LE_Married 26 13777 -1.94 0.43 -2 -2.00
## LE_MomDied 27 13777 -1.97 0.31 -2 -2.00
## LE_MoveIn 28 13777 -1.95 0.36 -2 -2.00
## LE_PartDied 29 13777 -1.99 0.18 -2 -2.00
## LE_SepPart 30 13777 -1.95 0.36 -2 -2.00
## Psych_LifeSat 31 13777 7.23 1.78 8 7.40
## LE_NewPart 32 13777 -1.91 0.52 -2 -2.00
## mad min max range skew kurtosis se
## Year* 0.00 2014 2014 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 17986163.38 901 35032702 35031801 0.03 -1.70 108737.47
## PROC_household 1795950.48 94 3343277 3343183 0.03 -1.70 10863.44
## Dem_DOB 23.72 -5 1996 2001 -2.10 2.40 5.79
## Dem_Sex 0.00 1 2 1 -0.15 -1.98 0.00
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -2 1 3 6.94 46.23 0.00
## LE_ChldMvOut 0.00 -2 1 3 5.82 31.86 0.00
## LE_Divorce 0.00 -2 1 3 14.23 200.60 0.00
## LE_DadDied 0.00 -2 1 3 9.12 81.11 0.00
## LE_Married 0.00 -2 1 3 6.67 42.52 0.00
## LE_MomDied 0.00 -2 1 3 9.56 89.36 0.00
## LE_MoveIn 0.00 -2 1 3 7.97 61.56 0.00
## LE_PartDied 0.00 -2 1 3 16.34 265.10 0.00
## LE_SepPart 0.00 -2 1 3 7.97 61.56 0.00
## Psych_LifeSat 1.48 -1 10 11 -1.17 2.12 0.02
## LE_NewPart 0.00 -2 1 3 5.41 27.29 0.00
## --------------------------------------------------------
## group: 2015
## vars n mean sd median trimmed
## Year* 1 13606 2015.00 0.00 2015 2015.00
## file* 2 13606 NaN NA NA NaN
## PROC_SID 3 13606 17266633.00 12967650.82 20248652 17290122.17
## PROC_household 4 13606 1726650.51 1295611.58 2024600 1731291.47
## Dem_DOB 5 13606 1841.06 481.62 1966 1964.70
## Dem_Sex 6 13606 1.12 1.67 2 1.51
## BF_C1 7 0 NaN NA NA NaN
## BF_E1 8 0 NaN NA NA NaN
## BF_A1 9 0 NaN NA NA NaN
## BF_O1 10 0 NaN NA NA NaN
## BF_N1 11 0 NaN NA NA NaN
## BF_A2 12 0 NaN NA NA NaN
## BF_C2 13 0 NaN NA NA NaN
## BF_E2 14 0 NaN NA NA NaN
## BF_O2 15 0 NaN NA NA NaN
## BF_N2 16 0 NaN NA NA NaN
## BF_C3 17 0 NaN NA NA NaN
## BF_E3 18 0 NaN NA NA NaN
## BF_A3 19 0 NaN NA NA NaN
## BF_O3 20 0 NaN NA NA NaN
## BF_N3 21 0 NaN NA NA NaN
## LE_ChldBrth 22 13606 -2.13 0.85 -2 -2.00
## LE_ChldMvOut 23 13606 -2.11 0.90 -2 -2.00
## LE_Divorce 24 13606 -2.18 0.76 -2 -2.00
## LE_DadDied 25 13606 -2.16 0.80 -2 -2.00
## LE_Married 26 13606 -2.13 0.85 -2 -2.00
## LE_MomDied 27 13606 -2.16 0.80 -2 -2.00
## LE_MoveIn 28 13606 -2.15 0.82 -2 -2.00
## LE_PartDied 29 13606 -2.18 0.76 -2 -2.00
## LE_SepPart 30 13606 -2.15 0.83 -2 -2.00
## Psych_LifeSat 31 13606 7.38 1.77 8 7.56
## LE_NewPart 32 13606 -2.09 0.93 -2 -2.00
## mad min max range skew kurtosis se
## Year* 0.00 2015 2015 0 NaN NaN 0.00
## file* NA Inf -Inf -Inf NA NA NA
## PROC_SID 18155623.08 901 35031602 35030701 -0.05 -1.67 111172.17
## PROC_household 1812236.84 94 3499900 3499806 -0.06 -1.67 11107.33
## Dem_DOB 20.76 -5 1997 2002 -3.57 10.74 4.13
## Dem_Sex 0.00 -5 2 7 -3.08 8.60 0.01
## BF_C1 NA Inf -Inf -Inf NA NA NA
## BF_E1 NA Inf -Inf -Inf NA NA NA
## BF_A1 NA Inf -Inf -Inf NA NA NA
## BF_O1 NA Inf -Inf -Inf NA NA NA
## BF_N1 NA Inf -Inf -Inf NA NA NA
## BF_A2 NA Inf -Inf -Inf NA NA NA
## BF_C2 NA Inf -Inf -Inf NA NA NA
## BF_E2 NA Inf -Inf -Inf NA NA NA
## BF_O2 NA Inf -Inf -Inf NA NA NA
## BF_N2 NA Inf -Inf -Inf NA NA NA
## BF_C3 NA Inf -Inf -Inf NA NA NA
## BF_E3 NA Inf -Inf -Inf NA NA NA
## BF_A3 NA Inf -Inf -Inf NA NA NA
## BF_O3 NA Inf -Inf -Inf NA NA NA
## BF_N3 NA Inf -Inf -Inf NA NA NA
## LE_ChldBrth 0.00 -5 1 6 -1.48 8.64 0.01
## LE_ChldMvOut 0.00 -5 1 6 -0.94 7.65 0.01
## LE_Divorce 0.00 -5 1 6 -2.86 10.27 0.01
## LE_DadDied 0.00 -5 1 6 -2.19 9.60 0.01
## LE_Married 0.00 -5 1 6 -1.47 8.62 0.01
## LE_MomDied 0.00 -5 1 6 -2.21 9.63 0.01
## LE_MoveIn 0.00 -5 1 6 -1.94 9.30 0.01
## LE_PartDied 0.00 -5 1 6 -2.99 10.37 0.01
## LE_SepPart 0.00 -5 1 6 -1.79 9.09 0.01
## Psych_LifeSat 1.48 -1 10 11 -1.26 2.42 0.02
## LE_NewPart 0.00 -5 1 6 -0.71 7.13 0.01Check Missings
How are missings coded in this data set? Do we need to make any changes to how they are coded?
# You should have noted some variables that needed "scrubbed" (changed to missing)
# change those to NA using your preferred method
dat <- dat %>% mutate_all(funs(mapvalues(., seq(-1,-7,-1), rep(NA,7), warn_missing = F)))Recode Variables
# You should have your keys. Reverse code the items that need reverse coded.
keys <- codebook$rev_code[codebook$rev_code == -1]
items <- codebook$new_name[codebook$rev_code == -1]
dat[,items] <- reverse.code(keys, dat[,items], mini = 1, maxi = 7)
# I'm going to give you this chunk because apparently some people don't know what year they were born
dat <- dat %>%
group_by(PROC_SID) %>%
mutate(
Dem_DOB = max(Dem_DOB, na.rm = T),
Dem_DOB = ifelse(is.infinite(Dem_DOB) == T, NA, Dem_DOB),
Dem_Sex = max(Dem_Sex, na.rm = T),
Dem_Sex = ifelse(is.infinite(Dem_Sex) == T, NA, Dem_Sex)
)Create New Variables
For these data, we need to create an age variable. There isn’t one in the data set.v
# create an age variable by subtracting the date of birth from 2005
# change gender to a factor
dat <- dat %>%
mutate(age = 2005 - Dem_DOB,
gender = factor(Dem_Sex, levels = c(1,2), labels = c("Male", "Female")))
# create a composite "parent died" variable
dat <- dat %>%
group_by(PROC_SID) %>%
mutate(LE_ParDied = max(LE_MomDied,LE_DadDied, na.rm = T),
LE_ParDied = ifelse(is.nan(LE_ParDied) == T, NA, LE_ParDied)) Create composites
For these data, we have lots of items, so we don’t just want to create composites for the Big 5, we also want to create composites for the facets of each of the Big 5. Use the methods we learned before to do so.
Personality
pers_dat <- dat %>%
gather(key = item, value = value, BF_A1:BF_O3, na.rm = T) %>%
separate(item, c("trait", "item"), -1) %>%
group_by(trait, PROC_SID, PROC_household, Year, Dem_DOB, age, Dem_Sex, gender) %>%
summarize(value = mean(value, na.rm = T)) %>%
spread(key = trait, value = value) %>%
ungroup() %>%
mutate(wave = as.numeric(Year) - 2005)Life Events
We want to code if someone experienced a life event anywhere within the study period. Experiment with different ways of doing this. Note that we want to figure out if a participant EVER responded with a 1 to any of our life event variables.
le_dat <- dat %>% select(-contains("BF")) %>%
gather(key = le, value = le_value, contains("LE_")) %>%
group_by(PROC_SID, age, gender, le) %>%
summarize(le_value = ifelse(any(le_value == 1), 1, 0),
le_value = ifelse(is.na(le_value), 0, le_value)) %>%
ungroup() %>%
spread(key = le, value = le_value)Merge the files
dat_final <- pers_dat %>%
full_join(le_dat) %>%
full_join(dat %>% filter(Year == 2005) %>% select(PROC_SID, Psych_LifeSat)) %>%
mutate(wave = as.numeric(Year) - 2005)Adding Predictors in MLM/HLM
Last week, we talked about building simple longitudinal models in R using MLM/HLM. This week, we’re going to extend the “simple” MLM growth curve model to include additional predictors. Depending on the type of predictor, you can add these predictors at level 1 or level 2. Figuring out which level your covariates are at is very important. Part of this will have to do with whether you expect level 1 and level 2 variables to interact.
To demonstrate what I mean by this, we can start by considering the case where we want to add a covariate like age at level 1, where we want to partial out how age influences the model, but not to see how personality changes over time as a function of age.
The simple growth model has the following form: Level 1: \(Y_{ij} = \beta_{0j} + \beta_{1j}*time_{ij} + \varepsilon_{ij}\)
Level 2:
Random Intercept: \(\beta_{0j} = \gamma_{00} + U_{0j}\)
Random Slope: \(\beta_{1j} = \gamma_{10} + U_{1j}\)
Additional Level 1 Covariate
When we add age as a covariate, it has the following form:
Level 1: \(Y_{ij} = \beta_{0j} + \beta_{1j}*time_{ij} + \beta_{2j}*age_{ij} + \varepsilon_{ij}\)
Level 2:
Intercept: \(\beta_{0j} = \gamma_{00} + U_{0j}\) (random intercept)
Time Slope: \(\beta_{1j} = \gamma_{10} + U_{1j}\) (random slope)
Age Slope: \(\beta_{2j} = \gamma_{20}\) (no random slope)
Note that this does add an additional level 2 equation, but we don’t add an addiitonal random slope. How does this affect our model? One of the easiest ways to see this is to “sub in” the level 2 models into the level 1 equation, which gives us:
\(Y_{ij} = \gamma_{00} + U_{0j} + (\gamma_{10} + U_{1j})*time_{ij} + (\gamma_{20})*age_{ij} + \varepsilon{ij}\)
So in this case, we are just adding one addition term (a fixed effect age term), which we would interpret the same as we would if we added an age variable in multiple regression.
Additional Level 2 Covariate
Sometimes, we are interested in covariates that we expect to interact with our level 1 covariates, but we have no reason to think that we should
Level 1: \(Y_{ij} = \beta_{0j} + \beta_{1j}*time_{ij} + \varepsilon_{ij}\)
Level 2:
Intercept: \(\beta_{0j} = \gamma_{00} + \gamma_{01}*life\_sat\_w1_{j} + U_{0j}\) (random intercept)
Time Slope: \(\beta_{1j} = \gamma_{10} + \gamma_{11}*life\_sat\_w1_{j} + U_{1j}\) (random slope)
Which combines to:
\(Y_{ij} = \gamma_{00} + \gamma{11}*life\_sat\_w1_{j} + U_{0j} + (\gamma_{10} + \gamma{11}*life\_sat\_w1_{j} + U_{1j})*time_{ij} + \varepsilon{ij}\)
\(Y_{ij} = \gamma_{00} + \gamma{11}*life\_sat\_w1_{j} + \gamma_{10}*time_{ij} + \gamma{11}*life\_sat\_w1_{j}*time_{ij} + U_{0j} + U_{1j}*time + \varepsilon_{ij}\)
Simple Growth Curve Model
But first, we’ll start by looking at a spaghetti plot of the personality data.
In R
(unconditional_model <- unique(dat_final %>% select(PROC_SID:BF_O, Psych_LifeSat, wave)) %>%
gather(key = Trait, value = value, BF_A:BF_O, na.rm = T) %>%
group_by(Trait, PROC_SID) %>%
mutate(n = n(), Year = as.numeric(Year)) %>% filter(n > 1) %>%
group_by(Trait) %>%
nest() %>%
mutate(model = map(data, ~lmer(value ~ 1 + (1|PROC_SID), data = .)),
ICC = map_dbl(model, reghelper::ICC),
summary = map(model, ~print(summary(.)))))## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ 1 + (1 | PROC_SID)
## Data: .
##
## REML criterion at convergence: 47367.7
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -6.0281 -0.4899 0.0211 0.5302 3.7581
##
## Random effects:
## Groups Name Variance Std.Dev.
## PROC_SID (Intercept) 0.1592 0.3990
## Residual 0.3759 0.6131
## Number of obs: 22093, groups: PROC_SID, 8563
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.732968 0.005996 789.4
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ 1 + (1 | PROC_SID)
## Data: .
##
## REML criterion at convergence: 52943.5
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.9661 -0.4748 -0.0339 0.4069 4.3265
##
## Random effects:
## Groups Name Variance Std.Dev.
## PROC_SID (Intercept) 0.2271 0.4766
## Residual 0.4733 0.6880
## Number of obs: 22057, groups: PROC_SID, 8552
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.056923 0.006963 582.7
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ 1 + (1 | PROC_SID)
## Data: .
##
## REML criterion at convergence: 55839
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -5.1119 -0.5818 -0.0509 0.6079 3.4235
##
## Random effects:
## Groups Name Variance Std.Dev.
## PROC_SID (Intercept) 0.2276 0.4771
## Residual 0.5565 0.7460
## Number of obs: 22076, groups: PROC_SID, 8558
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.584514 0.007232 633.9
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ 1 + (1 | PROC_SID)
## Data: .
##
## REML criterion at convergence: 72501.6
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.5972 -0.5295 -0.0141 0.6243 3.4435
##
## Random effects:
## Groups Name Variance Std.Dev.
## PROC_SID (Intercept) 0.9606 0.9801
## Residual 0.9541 0.9768
## Number of obs: 22080, groups: PROC_SID, 8560
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.05188 0.01252 323.7
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ 1 + (1 | PROC_SID)
## Data: .
##
## REML criterion at convergence: 64453.4
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.7680 -0.5165 0.0257 0.5374 3.7288
##
## Random effects:
## Groups Name Variance Std.Dev.
## PROC_SID (Intercept) 0.8265 0.9091
## Residual 0.6079 0.7797
## Number of obs: 22076, groups: PROC_SID, 8559
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.47130 0.01118 400# Thus, we see that we have due cause to use an HLM -- the ICC of the unconditional models was `r round((unconditional_model %>% filter(Trait == "BF_A"))$ICC,2)` (A), `r round((unconditional_model %>% filter(Trait == "BF_C"))$ICC,2)` (C), `r round((unconditional_model %>% filter(Trait == "BF_E"))$ICC,2)` (E), `r round((unconditional_model %>% filter(Trait == "BF_N"))$ICC,2)` (N), and `r round((unconditional_model %>% filter(Trait == "BF_O"))$ICC,2)` (O). Like last week, we can break down this output into its component parts:
Fixed Effects
Start by looking at the part that says “Fixed effects”. This will be most familiar to you and easy to interpret. These are simply the population level (average) effects. In this case we have:
In this case, for the fixed effects, we only have an intercept because this is the unconditional model. The intercept here is just the average of the outcome variable.
Random Effects
The random effects are your individual differences. This section will give you variance estimates of the level 1 and 2 (random effect) variances.
Because this is the unconditional model, we have only a random intercept variance, which tells us how much people differ in average levels of the outcome. Then, we would typically look at the ICC to see if we are capturing a fair amount of the variability in the outcome by estimating something separately for each person.
The estimates in the “Residual” section are the Level 1 residuals. So this would be found by extracting the residuals of the model (e.g. using the resid() function) and finding their variance. Ideally this value is small, which indicates that most people don’t differ much from the model.
Scaled Residuals
The scaled residuals can tell you about how your level 1 residuals are distributed. Remember that we want these to be normally distributed (in an ideal world). You can check these out and see if the estimates seem symmetrical. This isn’t a section of primary importance, but it’s good to check out.
Conditional Models: Adding Predictors
Let’s see if we can better predict participants’ change in sensation seeking over time by adding covariates.
| Predictor | Continuous | Categorical |
|---|---|---|
| Time Invariant | Life Satisfaction in 2005 | Life Event Group |
Time Invariant Predictors
Time Invariant Predictors: Continuous
As a reminder, the model we plan to fit is as follows:
Level 1: \(Y_{ij} = \beta_{0j} + \beta_{1j}*time_{ij} + \varepsilon_{ij}\)
Level 2:
Intercept: \(\beta_{0j} = \gamma_{00} + \gamma_{01}*life\_sat\_w1_{j} + U_{0j}\) (random intercept)
Time Slope: \(\beta_{1j} = \gamma_{10} + \gamma_{11}*life\_sat\_w1_{j} + U_{1j}\) (random slope)
But we’ll also need to test whether we need the random slope.
Continuous Example - Life Satisfaction
# no random slope
unconditional_model <- unconditional_model %>%
mutate(mod1 = map(data, ~lmer(value ~ wave + Psych_LifeSat + wave*Psych_LifeSat +
(1|PROC_SID), data = .)),
summary1 = map(mod1, ~print(summary(.))))Now, let’s add a random slope and compare the models. A couple of these will fail to converge, but we’ll get to how to deal with that later.
# random slope and intercept
unconditional_model <- unconditional_model %>%
mutate(mod2 = map(data, ~lmer(value ~ wave + Psych_LifeSat + wave*Psych_LifeSat +
(Year | PROC_SID), data = .)),
summary2 = map(mod2, ~print(summary(.))))## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ wave + Psych_LifeSat + wave * Psych_LifeSat + (Year |
## PROC_SID)
## Data: .
##
## REML criterion at convergence: 42925.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -5.9706 -0.5173 0.0185 0.5274 3.7962
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## PROC_SID (Intercept) 4.245e-01 0.6515277
## Year 2.729e-07 0.0005224 -1.00
## Residual 3.769e-01 0.6139485
## Number of obs: 20020, groups: PROC_SID, 7529
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.531765 0.032253 140.51
## wave 0.006612 0.005748 1.15
## Psych_LifeSat 0.031108 0.004445 7.00
## wave:Psych_LifeSat -0.001503 0.000790 -1.90
##
## Correlation of Fixed Effects:
## (Intr) wave Psy_LS
## wave -0.605
## Psych_LifSt -0.969 0.586
## wv:Psych_LS 0.588 -0.969 -0.606
## convergence code: 0
## Model failed to converge with max|grad| = 1.47439 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
##
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ wave + Psych_LifeSat + wave * Psych_LifeSat + (Year |
## PROC_SID)
## Data: .
##
## REML criterion at convergence: 47867.1
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.9157 -0.5001 -0.0344 0.4004 4.3180
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## PROC_SID (Intercept) 5.378e-01 0.7333207
## Year 3.551e-07 0.0005959 -1.00
## Residual 4.775e-01 0.6910110
## Number of obs: 19990, groups: PROC_SID, 7521
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 3.7948774 0.0367534 103.25
## wave -0.0012968 0.0064868 -0.20
## Psych_LifeSat 0.0325977 0.0050649 6.44
## wave:Psych_LifeSat 0.0009043 0.0008913 1.01
##
## Correlation of Fixed Effects:
## (Intr) wave Psy_LS
## wave -0.598
## Psych_LifSt -0.969 0.579
## wv:Psych_LS 0.580 -0.969 -0.598
## convergence code: 0
## Model failed to converge with max|grad| = 2.02804 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
##
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ wave + Psych_LifeSat + wave * Psych_LifeSat + (Year |
## PROC_SID)
## Data: .
##
## REML criterion at convergence: 50653.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -5.0778 -0.5595 -0.0230 0.5773 3.4507
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## PROC_SID (Intercept) 6.298e-01 0.7936125
## Year 3.998e-07 0.0006323 -1.00
## Residual 5.593e-01 0.7478441
## Number of obs: 20005, groups: PROC_SID, 7525
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.2805461 0.0390846 109.52
## wave 0.0130628 0.0070022 1.87
## Psych_LifeSat 0.0430005 0.0053864 7.98
## wave:Psych_LifeSat -0.0016047 0.0009623 -1.67
##
## Correlation of Fixed Effects:
## (Intr) wave Psy_LS
## wave -0.609
## Psych_LifSt -0.969 0.589
## wv:Psych_LS 0.591 -0.969 -0.609
## convergence code: 0
## Model failed to converge with max|grad| = 1.80725 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?
##
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ wave + Psych_LifeSat + wave * Psych_LifeSat + (Year |
## PROC_SID)
## Data: .
##
## REML criterion at convergence: 64962.4
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.8704 -0.5723 -0.0047 0.5708 3.5171
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## PROC_SID (Intercept) 1.071e+00 1.0347277
## Year 8.870e-07 0.0009418 -0.97
## Residual 9.475e-01 0.9733743
## Number of obs: 20009, groups: PROC_SID, 7527
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 5.480993 0.060318 90.87
## wave -0.069851 0.009187 -7.60
## Psych_LifeSat -0.185958 0.008314 -22.37
## wave:Psych_LifeSat 0.005297 0.001263 4.20
##
## Correlation of Fixed Effects:
## (Intr) wave Psy_LS
## wave -0.511
## Psych_LifSt -0.969 0.495
## wv:Psych_LS 0.496 -0.969 -0.512
## convergence code: 0
## unable to evaluate scaled gradient
## Model failed to converge: degenerate Hessian with 1 negative eigenvalues
##
## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ wave + Psych_LifeSat + wave * Psych_LifeSat + (Year |
## PROC_SID)
## Data: .
##
## REML criterion at convergence: 58230.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -4.6640 -0.5257 0.0279 0.5474 3.7185
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## PROC_SID (Intercept) 6.905e-01 0.8309479
## Year 7.424e-07 0.0008616 -1.00
## Residual 6.111e-01 0.7817529
## Number of obs: 20005, groups: PROC_SID, 7526
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 3.8504678 0.0537350 71.66
## wave -0.0004291 0.0074036 -0.06
## Psych_LifeSat 0.0928792 0.0074064 12.54
## wave:Psych_LifeSat -0.0013481 0.0010172 -1.33
##
## Correlation of Fixed Effects:
## (Intr) wave Psy_LS
## wave -0.459
## Psych_LifSt -0.969 0.445
## wv:Psych_LS 0.446 -0.969 -0.460
## convergence code: 0
## unable to evaluate scaled gradient
## Model failed to converge: degenerate Hessian with 1 negative eigenvaluesAs before, let’s unpack the results of one of the models (Agreeableness and Child Birth):
unconditional_model$summary2[[1]]## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ wave + Psych_LifeSat + wave * Psych_LifeSat + (Year |
## PROC_SID)
## Data: .
##
## REML criterion at convergence: 42925.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -5.9706 -0.5173 0.0185 0.5274 3.7962
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## PROC_SID (Intercept) 4.245e-01 0.6515277
## Year 2.729e-07 0.0005224 -1.00
## Residual 3.769e-01 0.6139485
## Number of obs: 20020, groups: PROC_SID, 7529
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.531765 0.032253 140.51
## wave 0.006612 0.005748 1.15
## Psych_LifeSat 0.031108 0.004445 7.00
## wave:Psych_LifeSat -0.001503 0.000790 -1.90
##
## Correlation of Fixed Effects:
## (Intr) wave Psy_LS
## wave -0.605
## Psych_LifSt -0.969 0.586
## wv:Psych_LS 0.588 -0.969 -0.606
## convergence code: 0
## Model failed to converge with max|grad| = 1.47439 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?Fixed Effects
Start by looking at the part that says “Fixed effects”. This will be most familiar to you and easy to interpret. These are simply the population level (average) effects. In this case we have:
wave: The average change in personality for each new wave of data was .007.
wave:Psych_LifeSat: Agreeableness doesn’t appear to change as a function of both wave and life satisfaction.
Random Effects
The random effects are your individual differences. This section will give you variance estimates of the level 1 and 2 (random effect) variances.
The estimates in the “Groups” PROC_SID are the level 2 random effects. In other words, if we were to extract the level 1 residuals or the level 2 random effects (\(r_{0j}\) and \(r_{1j}\)) and find their variance (old school style, like you learned in intro stats), these would be your variance estimates. The variance of the random intercepts is \(\tau_{00}^2\) (0 = intercept), while the variance of the random slopes is \(\tau_{11}^2\) (1 = slope). This also gives us a standard deviation, which if you remember, tells us something about the precision of the estimate.
The estimates in the “Residual” section are the Level 1 residuals. So this would be found by extracting the residuals of the model (e.g. using the resid() function) and finding their variance. Ideally this value is small, which indicates that most people don’t differ much from the model.
The only column we haven’t discussed in the “Corr” column. In our model, we (implicitly) specified that we not only wanted to model the level 2 variances but also the correlation between level 2 variables, which in this case, means the correlation between random slopes and intercepts. A positive correlation would mean that people with higher scores at baseline tended to show increases in personality over time, while a negative correlation would mean that people with lower scores at baseline tended to show increases. Sometimes we might want to remove this correlation, but that’s a topic for a later time. In this case, we see that the variables are almost perfectly correlated. That is likely a scaling issue that led to the non-convergence. There are ways to fix this (e.g. if we grand mean centered Psych_LifeSat. I encourage you to rerun the above with a centered life satisfaction variable if you’re curious. For now, this document will be long enough).
Scaled Residuals
The scaled residuals can tell you about how your level 1 residuals are distributed. Remember that we want these to be normally distributed (in an ideal world). You can check these out and see if the estimates seem symmetrical. This isn’t a section of primary importance, but it’s good to check out.
Correlation of Fixed Effects:
(I think this is the corrected correlation between wave and the outcome?)
Plots
We will plot these very similar to how we plotted moderated effects before. The only difference will be that we need to add the argument re.form = NA to the predict function.
Look at the graphs, what do you think this tells us about the relationship between personality, personality change, and life satisfaction?
fixed.frame <- unique(dat_final %>% select(PROC_SID, Psych_LifeSat)) %>%
summarise(mean = mean(Psych_LifeSat, na.rm = T),
sd = sd(Psych_LifeSat, na.rm = T))
pred_fun <- function(fit){
crossing(
wave = seq(0,8,1),
Psych_LifeSat = c(fixed.frame$mean-fixed.frame$sd,
fixed.frame$mean,
fixed.frame$mean+fixed.frame$sd)
) %>%
mutate(pred = predict(fit, newdata = ., re.form = NA))
}
unconditional_model <- unconditional_model %>%
mutate(pred1 = map(mod1, pred_fun),
pred2 = map(mod2, pred_fun))
unconditional_model %>% unnest(pred1) %>%
mutate(group = mapvalues(Psych_LifeSat, unique(Psych_LifeSat), c("-1SD", "0SD", "1SD"))) %>%
ggplot(aes(x = wave, y = pred, color = group)) +
geom_line() +
facet_wrap(~Trait, nrow = 2) +
theme_classic()
Tabling Results
The function below is the same one from before. I just included it again for reference.
table_fun <- function(model){
fixed <- broom::tidy(model) %>% filter(group == "fixed") %>%
select(term, estimate)
## add random effects ##
rand <- VarCorr(model)[[1]]
if(nrow(rand) > 1){
rand <- rand[1:nrow(rand), 1:nrow(rand)]
}
colnames(rand)[colnames(rand) == "(Intercept)"] <- "Intercept"
rownames(rand)[rownames(rand) == "(Intercept)"] <- "Intercept"
vars <- rownames(rand)
rand[upper.tri(rand)] <- NA
rand <- data.frame(rand) %>% mutate(var1 = rownames(.)) %>%
gather(key = var2, value = estimate, -var1, na.rm = T) %>%
mutate(var1 = mapvalues(var1, vars, 0:(length(vars)-1)),
var2 = mapvalues(var2, unique(var2), 0:(length(vars)-1))) %>%
filter(var1 == var2) %>%
unite(var, var1, var2, sep = "") %>%
mutate(var = sprintf("$\\tau_{%s}$", var))
## get confidence intervals ##
CI <- data.frame(confint.merMod(model, method = "boot", nsim = 10, oldNames = F)) %>%
mutate(term = rownames(.)) %>% setNames(c("lower", "upper", "term"))
CI %>% filter(term == "sigma") %>%
mutate(estimate = sigma(model),
term = "$\\sigma^2$",
type = "Residuals")
## Get ICC & R2 values ##
R2 <- MuMIn::r.squaredGLMM(model)
## format the fixed effects
fixed <- fixed %>% left_join(CI %>% filter(!grepl(".sig", term))) %>%
mutate(type = "Fixed Parts", term = str_remove_all(term, "[()]"))
rand <- rand %>%
left_join(
CI %>% filter(grepl("sd", term)) %>%
mutate(lower = lower^2, upper = upper^2,
var = mapvalues(term, unique(term), 0:(length(unique(term))-1)),
var = sprintf("$\\tau_{%s%s}$", var, var)) %>% select(-term)) %>%
mutate(type = "Random Parts") %>% rename(term = var)
mod_terms <- tribble(
~term, ~estimate, ~type,
# "ICC", ICC, "Model Terms",
"$R^2_m$", R2[1], "Model Terms",
"$R^2_c$", R2[2], "Model Terms"
)
tab <- fixed %>%
full_join(rand) %>%
mutate(CI = sprintf("[%.2f, %.2f]", lower, upper)) %>%
select(-lower, -upper) %>%
full_join(mod_terms) %>%
mutate(estimate = sprintf("%.2f", estimate)) %>%
dplyr::rename(b = estimate) %>%
select(type, everything())
return(tab)
}We can use this function for the models in our table:
unconditional_model <- unconditional_model %>% mutate(l.tab2 = map(mod2, table_fun))
unconditional_model %>% unnest(l.tab2)This function does most of the heavy lifting for us, but we will need to reorganize. With 5 traits, I usually do something like the following:
(tab <- unconditional_model %>%
# unnest the table results
unnest(l.tab2) %>%
# remove the ICC
select(-ICC) %>%
# make the b and CI long for reshaping
gather(key = est, value = value, b, CI, na.rm = T) %>%
# create a new variable joining together the type of term (b, CI) and trait
unite(tmp, Trait, est, sep = ".") %>%
# change to wide format
spread(key = tmp, value = value) %>%
# factor the rows to control rearrangement
mutate(type = factor(type, levels = c("Fixed Parts", "Random Parts", "Model Terms"))) %>%
# rearrange the variables
arrange(type))Then, we can use this with knitr::kable() and the kableExtra package to create a nicely formatted table output.
tab %>%
knitr::kable(., "html", booktabs = T, escape = F,
# set the column names
col.names = c("", "Term", rep(c("b", "CI"), times = 5)),
# set column alignment
align = c("l", "l", rep("c", 10))) %>%
kable_styling(full_width = F) %>%
# create floating header above
add_header_above(c(" " = 2, "Agreeableness" = 2, "Conscientiousness" = 2,
"Extraversion" = 2, "Neuroticism" = 2, "Openness" = 2))| Term | b | CI | b | CI | b | CI | b | CI | b | CI | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Fixed Parts | Intercept | 4.53 | [4.48, 4.59] | 3.79 | [3.73, 3.84] | 4.28 | [4.21, 4.35] | 5.48 | [5.39, 5.55] | 3.85 | [3.73, 3.89] |
| Fixed Parts | Psych_LifeSat | 0.03 | [0.02, 0.04] | 0.03 | [0.03, 0.04] | 0.04 | [0.04, 0.05] | -0.19 | [-0.20, -0.17] | 0.09 | [0.09, 0.11] |
| Fixed Parts | wave | 0.01 | [0.00, 0.02] | -0.00 | [-0.01, 0.01] | 0.01 | [0.01, 0.03] | -0.07 | [-0.08, -0.07] | -0.00 | [-0.01, 0.01] |
| Fixed Parts | wave:Psych_LifeSat | -0.00 | [-0.00, -0.00] | 0.00 | [-0.00, 0.00] | -0.00 | [-0.00, -0.00] | 0.01 | [0.00, 0.01] | -0.00 | [-0.00, 0.00] |
| Random Parts | \(\tau_{00}\) | 0.42 | [0.42, 0.43] | 0.54 | [0.53, 0.54] | 0.63 | [0.62, 0.64] | 1.07 | [1.20, 1.36] | 0.69 | [0.77, 0.81] |
| Random Parts | \(\tau_{11}\) | 0.00 | [0.00, 0.00] | 0.00 | [0.00, 0.00] | 0.00 | [0.00, 0.00] | 0.00 | [0.00, 0.00] | 0.00 | [0.00, 0.00] |
| Model Terms | \(R^2_c\) | 0.30 | NA | 0.31 | NA | 0.29 | NA | 0.51 | NA | 0.58 | NA |
| Model Terms | \(R^2_m\) | 0.00 | NA | 0.01 | NA | 0.01 | NA | 0.05 | NA | 0.02 | NA |
Time Invariant Predictors: Categorical
Categorical Example - 2 level group
In the case of categorical moderators, it’s slightly easier to understand because the different levels at which thee may be different relationships among predictors and outcomes.
In the case of life events moderating personality change, the model we plan to fit is as follows:
Level 1: \(Y_{ij} = \beta_{0j} + \beta_{1j}*time_{ij} + \varepsilon_{ij}\)
Level 2:
Intercept: \(\beta_{0j} = \gamma_{00} + \gamma_{01}*life\_event_{j} + U_{0j}\) (random intercept)
Time Slope: \(\beta_{1j} = \gamma_{10} + \gamma_{11}*life\_event_w1_{j} + U_{1j}\) (random slope)
Where \(life\_event_{j}\) is a dummy coded variable that is coded 1 when someone experienced a life event between 2005 and 2015, and 0 otherwise.
And plot it.
And model it:
nested.mods <- dat_final %>%
gather(key = Trait, value = value, BF_A:BF_O, na.rm = T) %>%
gather(key = Event, value = le_value, contains("LE"), na.rm = T) %>%
mutate(le_value = factor(le_value), Year = as.numeric(Year)) %>%
group_by(Trait, Event, PROC_SID) %>%
mutate(n = n()) %>% filter(n > 1) %>%
filter(!is.na(value) & !is.na(le_value)) %>%
group_by(Trait, Event) %>%
nest()
# no random slope
nested.mods <- nested.mods %>%
mutate(mod1 = map(data, ~lmer(value ~ wave + le_value + wave:le_value +
(1|PROC_SID), data = .)),
summary1 = map(mod1, ~summary(.)))
# add a random slope
nested.mods <- nested.mods %>%
mutate(mod2 = map(data, ~lmer(value ~ wave + le_value + wave:le_value +
(Year|PROC_SID), data = .)),
summary2 = map(mod2, ~summary(.)))
(nested.mods %>%
mutate(comp = map2(mod1, mod2, anova)))$comp[[1]]As before, let’s unpack the results of one of the models (Agreeableness and Child Birth):
nested.mods$summary2[[1]]## Linear mixed model fit by REML ['lmerMod']
## Formula: value ~ wave + le_value + wave:le_value + (Year | PROC_SID)
## Data: .
##
## REML criterion at convergence: 47381.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -5.9907 -0.5136 0.0233 0.5264 3.7713
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## PROC_SID (Intercept) 4.231e-01 0.6504513
## Year 2.730e-07 0.0005225 -1.00
## Residual 3.757e-01 0.6129192
## Number of obs: 22093, groups: PROC_SID, 8563
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 4.743753 0.008301 571.4
## wave -0.004116 0.001448 -2.8
## le_value1 0.030863 0.024441 1.3
## wave:le_value1 0.001642 0.004183 0.4
##
## Correlation of Fixed Effects:
## (Intr) wave le_vl1
## wave -0.639
## le_value1 -0.340 0.217
## wave:le_vl1 0.221 -0.346 -0.652
## convergence code: 0
## Model failed to converge with max|grad| = 1.37389 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
## - Rescale variables?
## Model is nearly unidentifiable: large eigenvalue ratio
## - Rescale variables?Fixed Effects
Start by looking at the part that says “Fixed effects”. This will be most familiar to you and easy to interpret. These are simply the population level (average) effects. In this case we have:
wave: The average change in personality for each new wave of data was .007.
wave:le_value: People who experience a life event don’t change differently than those who didn’t.
Random Effects
The random effects are your individual differences. This section will give you variance estimates of the level 1 and 2 (random effect) variances.
These are the same as we discussed before for the continuous moderator. You’ll notice that these estimates differ in magnitude. That’s because they are basically the individual differences in mean level personality and personality change that can’t be accounted for by our predcitors.
Scaled Residuals
The scaled residuals can tell you about how your level 1 residuals are distributed. Remember that we want these to be normally distributed (in an ideal world). You can check these out and see if the estimates seem symmetrical. This isn’t a section of primary importance, but it’s good to check out.
Correlation of Fixed Effects:
(I think this is the corrected correlation between wave and the outcome?)
Plots
pred_fun <- function(fit){
crossing(
wave = seq(0,8,1),
le_value = factor(c(0,1))
) %>%
mutate(pred = predict(fit, newdata = ., re.form = NA))
}
nested.mods <- nested.mods %>%
mutate(pred1 = map(mod1, pred_fun),
pred2 = map(mod2, pred_fun))
nested.mods %>% unnest(pred1) %>%
ggplot(aes(x = wave, y = pred, color = le_value)) +
geom_line() +
facet_grid(Event~Trait) +
theme_classic()
Tabling Results
nested.mods <- nested.mods %>% mutate(l.tab2 = map(mod2, table_fun))
nested.mods %>% unnest(l.tab2)This function does most of the heavy lifting for us, but we will need to reorganize. With 5 traits, I usually do something like the following:
(tab <- nested.mods %>%
# unnest the table results
unnest(l.tab2) %>%
# make the b and CI long for reshaping
gather(key = est, value = value, b, CI, na.rm = T) %>%
# create a new variable joining together the type of term (b, CI) and trait
unite(tmp, Trait, est, sep = ".") %>%
# change to wide format
spread(key = tmp, value = value) %>%
# factor the rows to control rearrangement
mutate(type = factor(type, levels = c("Fixed Parts", "Random Parts", "Model Terms"))) %>%
# rearrange the variables
arrange(type, term))Then, we can use this with knitr::kable() and the kableExtra package to create a nicely formatted table output.
tab %>%
filter(type == "Fixed Parts") %>%
knitr::kable(., "html", booktabs = T, escape = F,
# set the column names
col.names = c("", "Event", "Term", rep(c("b", "CI"), times = 5)),
# set column alignment
align = c("l", "l", rep("c", 10))) %>%
kable_styling(full_width = F) %>%
# create floating header above
add_header_above(c(" " = 3, "Agreeableness" = 2, "Conscientiousness" = 2,
"Extraversion" = 2, "Neuroticism" = 2, "Openness" = 2))| Event | Term | b | CI | b | CI | b | CI | b | CI | b | CI | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LE_ChldBrth | Fixed Parts | Intercept | 4.74 | [4.74, 4.76] | 4.02 | [4.00, 4.04] | 4.58 | [4.57, 4.60] | 4.19 | [4.16, 4.21] | 4.50 | [4.47, 4.51] |
| LE_ChldMvOut | Fixed Parts | Intercept | 4.74 | [4.72, 4.75] | 4.05 | [4.04, 4.07] | 4.59 | [4.58, 4.60] | 4.16 | [4.13, 4.19] | 4.49 | [4.47, 4.51] |
| LE_DadDied | Fixed Parts | Intercept | 4.75 | [4.75, 4.78] | 4.03 | [4.02, 4.05] | 4.59 | [4.57, 4.61] | 4.16 | [4.12, 4.22] | 4.50 | [4.48, 4.51] |
| LE_Divorce | Fixed Parts | Intercept | 4.75 | [4.73, 4.76] | 4.03 | [4.02, 4.05] | 4.58 | [4.58, 4.60] | 4.16 | [4.14, 4.19] | 4.50 | [4.49, 4.52] |
| LE_Married | Fixed Parts | Intercept | 4.74 | [4.73, 4.75] | 4.01 | [4.00, 4.03] | 4.59 | [4.57, 4.60] | 4.18 | [4.16, 4.20] | 4.49 | [4.48, 4.51] |
| LE_MomDied | Fixed Parts | Intercept | 4.74 | [4.73, 4.75] | 4.04 | [4.02, 4.06] | 4.58 | [4.57, 4.60] | 4.16 | [4.14, 4.19] | 4.50 | [4.49, 4.51] |
| LE_MoveIn | Fixed Parts | Intercept | 4.75 | [4.73, 4.76] | 4.01 | [4.01, 4.03] | 4.58 | [4.57, 4.60] | 4.17 | [4.13, 4.20] | 4.49 | [4.48, 4.51] |
| LE_NewPart | Fixed Parts | Intercept | 4.75 | [4.73, 4.76] | 4.02 | [4.01, 4.03] | 4.58 | [4.57, 4.60] | 4.17 | [4.16, 4.20] | 4.49 | [4.47, 4.51] |
| LE_ParDied | Fixed Parts | Intercept | 4.74 | [4.73, 4.76] | 4.03 | [4.02, 4.05] | 4.59 | [4.57, 4.60] | 4.16 | [4.13, 4.18] | 4.49 | [4.48, 4.54] |
| LE_PartDied | Fixed Parts | Intercept | 4.75 | [4.74, 4.76] | 4.04 | [4.03, 4.06] | 4.58 | [4.57, 4.60] | 4.16 | [4.15, 4.19] | 4.51 | [4.49, 4.53] |
| LE_SepPart | Fixed Parts | Intercept | 4.75 | [4.73, 4.77] | 4.02 | [4.00, 4.03] | 4.59 | [4.58, 4.60] | 4.16 | [4.12, 4.18] | 4.49 | [4.48, 4.50] |
| LE_ChldBrth | Fixed Parts | le_value1 | 0.03 | [0.01, 0.07] | 0.16 | [0.10, 0.21] | -0.01 | [-0.08, 0.01] | -0.15 | [-0.21, -0.11] | 0.01 | [-0.06, 0.07] |
| LE_ChldMvOut | Fixed Parts | le_value1 | 0.05 | [0.02, 0.10] | -0.10 | [-0.13, -0.07] | -0.07 | [-0.11, -0.04] | 0.04 | [-0.02, 0.13] | 0.04 | [0.01, 0.06] |
| LE_DadDied | Fixed Parts | le_value1 | 0.02 | [-0.03, 0.07] | 0.07 | [0.03, 0.08] | -0.09 | [-0.13, -0.04] | 0.05 | [-0.07, 0.21] | 0.03 | [-0.01, 0.12] |
| LE_Divorce | Fixed Parts | le_value1 | -0.02 | [-0.13, 0.08] | 0.03 | [-0.02, 0.09] | -0.13 | [-0.20, -0.09] | 0.14 | [0.05, 0.28] | 0.04 | [-0.08, 0.15] |
| LE_Married | Fixed Parts | le_value1 | 0.07 | [0.03, 0.14] | 0.22 | [0.18, 0.26] | -0.07 | [-0.14, -0.02] | -0.11 | [-0.15, -0.02] | 0.13 | [0.06, 0.19] |
| LE_MomDied | Fixed Parts | le_value1 | 0.06 | [-0.01, 0.10] | -0.01 | [-0.07, 0.03] | -0.03 | [-0.10, 0.03] | 0.07 | [-0.03, 0.15] | 0.03 | [-0.03, 0.07] |
| LE_MoveIn | Fixed Parts | le_value1 | 0.02 | [-0.02, 0.05] | 0.24 | [0.19, 0.28] | -0.06 | [-0.08, -0.02] | 0.00 | [-0.08, 0.05] | 0.14 | [0.05, 0.18] |
| LE_NewPart | Fixed Parts | le_value1 | 0.02 | [-0.02, 0.06] | 0.27 | [0.22, 0.35] | -0.11 | [-0.16, -0.02] | -0.08 | [-0.16, 0.07] | 0.24 | [0.15, 0.34] |
| LE_ParDied | Fixed Parts | le_value1 | 0.05 | [0.02, 0.06] | 0.04 | [0.01, 0.07] | -0.06 | [-0.08, -0.04] | 0.05 | [-0.10, 0.09] | 0.04 | [0.01, 0.10] |
| LE_PartDied | Fixed Parts | le_value1 | -0.05 | [-0.10, 0.02] | -0.24 | [-0.32, -0.15] | 0.08 | [0.01, 0.18] | 0.25 | [0.18, 0.40] | -0.28 | [-0.40, -0.17] |
| LE_SepPart | Fixed Parts | le_value1 | 0.00 | [-0.03, 0.04] | 0.17 | [0.13, 0.25] | -0.13 | [-0.20, -0.10] | 0.09 | [0.02, 0.16] | 0.13 | [0.05, 0.16] |
| LE_ChldBrth | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.01] | -0.03 | [-0.03, -0.03] | -0.01 | [-0.01, -0.00] |
| LE_ChldMvOut | Fixed Parts | wave | -0.00 | [-0.01, 0.00] | 0.00 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.03, -0.03] | -0.01 | [-0.01, -0.00] |
| LE_DadDied | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.04, -0.03] | -0.01 | [-0.01, -0.00] |
| LE_Divorce | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.04, -0.03] | -0.01 | [-0.01, -0.01] |
| LE_Married | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [0.00, 0.01] | -0.03 | [-0.04, -0.03] | -0.01 | [-0.01, -0.01] |
| LE_MomDied | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.01] | -0.03 | [-0.03, -0.03] | -0.01 | [-0.01, -0.01] |
| LE_MoveIn | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.03, -0.03] | -0.01 | [-0.01, -0.01] |
| LE_NewPart | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.04, -0.03] | -0.01 | [-0.01, -0.00] |
| LE_ParDied | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.03, -0.02] | -0.01 | [-0.01, -0.01] |
| LE_PartDied | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.03, -0.03] | -0.01 | [-0.01, -0.01] |
| LE_SepPart | Fixed Parts | wave | -0.00 | [-0.01, -0.00] | 0.01 | [0.00, 0.01] | 0.00 | [-0.00, 0.00] | -0.03 | [-0.03, -0.03] | -0.01 | [-0.01, -0.01] |
| LE_ChldBrth | Fixed Parts | wave:le_value1 | 0.00 | [-0.00, 0.01] | -0.01 | [-0.02, -0.00] | -0.01 | [-0.02, -0.00] | 0.00 | [-0.01, 0.01] | 0.00 | [-0.00, 0.01] |
| LE_ChldMvOut | Fixed Parts | wave:le_value1 | -0.00 | [-0.01, 0.01] | 0.01 | [0.00, 0.02] | 0.00 | [-0.00, 0.01] | -0.00 | [-0.02, 0.01] | 0.00 | [-0.00, 0.01] |
| LE_DadDied | Fixed Parts | wave:le_value1 | -0.00 | [-0.01, 0.00] | 0.00 | [-0.00, 0.01] | 0.01 | [-0.00, 0.02] | 0.00 | [-0.01, 0.01] | -0.01 | [-0.03, -0.00] |
| LE_Divorce | Fixed Parts | wave:le_value1 | 0.01 | [-0.00, 0.02] | -0.00 | [-0.01, 0.01] | 0.01 | [0.00, 0.02] | -0.01 | [-0.02, 0.01] | 0.01 | [-0.00, 0.03] |
| LE_Married | Fixed Parts | wave:le_value1 | 0.00 | [-0.00, 0.02] | -0.01 | [-0.01, 0.00] | -0.00 | [-0.01, 0.01] | 0.01 | [-0.00, 0.02] | -0.00 | [-0.01, 0.01] |
| LE_MomDied | Fixed Parts | wave:le_value1 | -0.00 | [-0.01, 0.01] | -0.00 | [-0.01, 0.01] | 0.00 | [-0.00, 0.01] | -0.00 | [-0.01, 0.01] | 0.01 | [0.00, 0.02] |
| LE_MoveIn | Fixed Parts | wave:le_value1 | 0.01 | [-0.00, 0.01] | -0.00 | [-0.02, 0.01] | -0.00 | [-0.01, 0.00] | -0.00 | [-0.01, 0.01] | -0.00 | [-0.02, 0.01] |
| LE_NewPart | Fixed Parts | wave:le_value1 | 0.01 | [0.01, 0.01] | 0.01 | [0.00, 0.02] | 0.01 | [-0.00, 0.02] | 0.00 | [-0.01, 0.02] | -0.00 | [-0.02, 0.01] |
| LE_ParDied | Fixed Parts | wave:le_value1 | -0.00 | [-0.01, 0.00] | -0.00 | [-0.01, 0.01] | 0.01 | [0.00, 0.01] | -0.00 | [-0.01, 0.00] | 0.00 | [-0.01, 0.01] |
| LE_PartDied | Fixed Parts | wave:le_value1 | 0.01 | [-0.01, 0.02] | 0.01 | [-0.00, 0.02] | 0.00 | [-0.02, 0.01] | -0.01 | [-0.03, 0.01] | 0.01 | [0.01, 0.03] |
| LE_SepPart | Fixed Parts | wave:le_value1 | 0.01 | [0.00, 0.02] | 0.00 | [-0.00, 0.01] | 0.01 | [0.01, 0.02] | -0.02 | [-0.04, -0.00] | 0.01 | [0.00, 0.01] |
Random Effects
We often also want to plot random effects to look at individual level trajectories / variability. Again, we need to use the predict() function to get these values. You can also totally use matrix algebra if you’re like me and want to know what’s happening “under the hood.”
#### Categorical
ran_pred_fun <- function(fit){
tbl_df(fit@frame) %>%
mutate(pred = predict(fit, newdata = .))
}
nested.mods <- nested.mods %>%
mutate(ran_pred1 = map(mod1, ran_pred_fun),
ran_pred2 = map(mod2, ran_pred_fun))
sample_subs <- c(sample((dat_final %>% filter(LE_ChldBrth == 0))$PROC_SID, 100),
sample((dat_final %>% filter(LE_ChldBrth == 1))$PROC_SID, 100))
nested.mods %>% unnest(ran_pred2) %>%
filter(PROC_SID %in% sample_subs) %>%
ggplot(aes(x = wave, y = pred, group = PROC_SID, color = le_value)) +
geom_line(size = .5, alpha = .5) +
labs(x = "Year", y = "Predicted Personality Composite",
title = "Predicted Random Effect Trajectories") +
facet_grid(Event ~ Trait) +
theme_classic() +
theme(axis.text = element_text(face = "bold", size = rel(1.2)),
axis.title = element_text(face = "bold", size = rel(1.2)),
legend.title = element_text(face = "bold", size = rel(1.2)),
plot.title = element_text(face = "bold", size = rel(1.2), hjust = .5))
Continuous
sample_subs <- sample(unique(dat_final$PROC_SID), 200)
unconditional_model <- unconditional_model %>%
mutate(ran_pred1 = map(mod1, ran_pred_fun),
ran_pred2 = map(mod2, ran_pred_fun))
unconditional_model %>%
unnest(ran_pred2) %>%
filter(PROC_SID %in% sample_subs) %>%
ggplot(aes(x = wave, y = pred, group = PROC_SID, color = Psych_LifeSat)) +
geom_line(size = .5, alpha = .5) +
labs(x = "Age", y = "Predicted Personality Composite",
title = "Predicted Random Effect Trajectories") +
facet_wrap(~ Trait, nrow = 2) +
theme_classic() +
theme(axis.text = element_text(face = "bold", size = rel(1.2)),
axis.title = element_text(face = "bold", size = rel(1.2)),
legend.title = element_text(face = "bold", size = rel(1.2)),
plot.title = element_text(face = "bold", size = rel(1.2), hjust = .5))